The challenge is that species differences between mice/rats and humans frequently confound translation of animal studies of efficacy and drug safety to predicting outcomes in humans.

In a Challenges and Opportunities report issued by the US Food and Drug Administration (FDA), it was noted that the inability to predict lack of effectiveness and safety problems before human testing or early in clinical trials dramatically escalates costs of drug development.

According to the FDA, drug developers could save $100 million in development costs per drug with a 10% improvement in predicting failures before clinical trials1.

As a result there is a growing need for animal models that more faithfully recapitulate human physiology and disease pathology. Toward this goal, various methods have been developed to generate mice with closer biological similarities to humans. In this article, we review how mice have been engineered – either genetically or by tissue cell or tissue engraftment – to create what are known as ‘humanised’ precision research models. These models are quickly becoming an essential tool in drug discovery and development2-5, and we discuss the use of these models in several key application areas, including: Drug Metabolism and Pharmacokinetics (DMPK), carcinogenicity testing, neuroscience and immuno-oncology.

By leveraging precision research models to better model human disease pathology and/or predict human outcomes with regard to drug efficacy, safety and toxicity, the cost and time related to drug discovery and development has the potential to shrink significantly.

Improving prediction of human outcomes
Humanised precision models can be generated by genetically engineering a human genetic component into a mouse or a rat, or by engrafting human cells and tissue to mimic human-like organ systems. These models can serve as tools to model specific human disease pathology (eg, neurodegeneration seen in Alzheimer’s disease (AD), to test drug efficacy, or to better understand the safety profile of a drug in development. Together, the models provide critical data that are valuable at go/no-go decision points.

There is no single method used to generate these models, and the exact model used depends heavily on the question(s) being asked.

Some recent examples of precision research models demonstrate how they have helped overcome challenges commonly encountered by drug developers.

The advancement of genetically engineered models
More than 30 years have passed since the development of the first mouse model with a randomly inserted human transgene6. Today’s readily available genome editing technologies allow genomic replacement of mouse genes with their human counterpart with very high precision. This has resulted in the generation of complex models with the potential to revolutionise aspects of drug discovery and development before human testing even begins.

As a testament to the rapid advancement in the complexity of these models, Scheer et al recently reported on the generation of the most complex precision engineered research model to date. Thirty-three mouse genes were replaced with human genetic counterparts in a single mouse. The model, known as the hPXR/CAR/CYP3A4/ 2D6/2C9 mouse, had the mouse pregnane X receptor (Pxr) and constitutive androstane receptor (Car), along with the Cyp2c Cyp2d and Cyp3a gene clusters, exchanged with human PXR, CAR, and the CYP2C9, CYP2D6 and CYP3A4/7 loci, respectively7. A truly astonishing feat.

Utility in DMPK
It is a given that the variability in drug pharmacokinetics between mouse and man hinders prediction. As a result the ability to better predict drug metabolism and disposition in humans holds significant benefit to both lead candidate selection, and to prevent late-stage attrition of drug candidates. In recent years, the number of mouse models that have been engineered to possess more human-like drug metabolism and disposition has increased considerably8.

In general, many drug-drug interactions are caused by PXR-mediated transcriptional activation of CYP genes. While evaluation of PXR activation and downstream induction of proteins involved in drug elimination and distribution is an important activity in drug development, PXR-induced CYP 3A4 expression differs across species. This is mainly because of differences in nuclear receptor interaction with various drugs, which greatly limits the utility of animal models in the prediction of clinically relevant drug-drug interactions (DDIs) by PXR- or CAR-mediated induction of CYP3A4.

To address this limitation, Hasegawa et al9 generated a multiple humanised mouse model that replaced the mouse Pxr and Car, along with the Cyp3a gene cluster, with the human PXR and CAR counterparts, and a large human genomic region carrying CYP3A4 and CYP3A7, respectively. This humanised PXR/CAR/CYP3A4 mouse shows a human-like CYP3A4 induction response to different PXR activators, and provides an experimental approach to quantitatively predict clinically relevant drug-drug interactions in humans. Such a mouse can be extremely useful in evaluating CYP induction profiles and potential DDIs of compounds in development.

More recently, Scheer et al developed a more complex model for predicting human responses to drugs by exchanging 31 mouse P450 genes from the Cyp2c, Cyp2d, and Cyp3a gene families for their corresponding human counterparts (humanising the majority of Phase I drug metabolism pathway), in addition to PXR and CAR humanisation. According to the authors, this model is the most complex genetically humanised model reported to date, and the P450s function as predicted. The study shows the utility of this mouse in several applications including predicting induction and inhibition of CYPs and predicting DDIs7.

Membrane transporters can be major determinants of pharmacokinetic, safety, and efficacy profiles of drugs, and there is a clear need to understand which transporters are clinically important in drug disposition and metabolism10. Given this need, the International Transporter Consortium (ITC) has provided guidance on the most important drug transporters for investigation, and a group of for-profit and non-profit entities have collaborated to design, generate, characterise and validate genetically humanised drug transporter mice11. While these models are still being validated, they certainly hold tremendous promise to help accelerate drug development activities.

Utility in assessing human carcinogenic risk for novel pharmaceuticals
For decades, the two-year rodent carcinogenicity assay has been the regulatory standard for assessing human carcinogenic risk of novel pharmaceuticals12. As specified by ICH S1 guidelines, this would normally involve two long-term rodent carcinogenicity assays. In the late 1990s, the FDA approved the use of short, or medium-term bioassays in transgenic mice as an alternative carcinogenicity assay. Two transgenic models – TSG-p53® and rasH2 – became the most popular alternatives. While both models show increased susceptibility to tumour formation following carcinogen exposure, rasH2 is the most widespread model for short-term carcinogenicity assays. This is primarily due to rasH2 mice being responsive to compounds that exhibit both genotoxic and non-genotoxic mechanisms of action13-15.

The rasH2 model has a low incidence of spontaneous tumour formation up to six months of age, and these mice are highly susceptible to tumour development following carcinogen exposure16. Short-term rasH2 studies, performed over a six-month period (compared to the two-year lifetime assay) have been demonstrated to predict known and likely human carcinogens just as well as conventional two-year mouse lifetime assays, but with fewer false positives17-18.

The faster tumorigenic and more accurate response in these transgenic models translates to significantly shorter carcinogenicity studies, fewer animals required for study completion, and costs reductions to ~$1 million, as compared to >$2 million for the two-year study. Ultimately the transgenic mouse accelerates time-to-data acquisition and generates a more timely evaluation of the relevant human carcinogenic risk.

The initial adoption of this model for short-term carcinogenicity assays by the pharmaceutical industry was slow13 and this was primarily due to lack of historical control data, which are required for regulatory submissions. This is now changing as several CROs have generated historical data (many published) and with the FDA acceptance of this model’s utility, >40% of all carcinogenicity protocols (data obtained from FDA), and 75% of all mouse carcinogenicity studies15 submitted to the FDA use the rasH2 mouse for assessing human carcinogenic risk.

Over the past three years, significant changes have been considered at the international level to address whether the cost and time involved with two-year carcinogenicity assays is necessary. Based on retrospective analyses, an ICH S1 expert working group (S1 EWG) has proposed changes to the carcinogenicity testing guidelines that would perhaps eliminate the need for a two-year rat carcinogenicity bioassay if other sufficient data are available that warrant an exemption19. These proposed changes could apply to 30-40% of pharmaceuticals submitted for application18. In the context of the proposed ICH S1 guideline changes, models such as rasH2 have the potential to fulfill a greater role in the carcinogenicity testing process. Furthermore, the rasH2 mouse may also play an increased role in testing the carcinogenic potential of medical devices, as it does not exhibit the solid-state tumour response seen in two-year rodent bioassays when surgically implanted with biomaterials21-22.

Utility in neuroscience
A key challenge in the discovery of novel neuroscience drugs is a lack of good models that recapitulate human neuronal pathology. Such models are important for efficacy testing of drugs in development, and in particular, the Alzheimer’s Disease (AD) field is an area of high need for models that can recapitulate the neurodegenerative pathology and cognitive decline seen in human patients. Notably, some progress has been made in this area, and we describe one such model below.

The APPSWE mouse (aka Tg2576) has been used in preclinical studies for several of the approved AD drugs (eg, donepezil, galantamine, memantine). APPSWE mice accumulate beta-amyloid (Ab) plaques as they age and develop cognitive and behavioural deficits similar to those seen in human AD patients. These key characteristics make them a platform model for screening treatments that target Ab plaques and ameliorate the associated cognitive and behavioural deficits. A recent pre-clinical study at Merck used the APPSWE mouse to address a key challenge with the current standard care in AD. The majority of the drugs approved by the FDA for treatment of cognitive symptoms in AD patients are cholinesterase inhibitors and include drugs such as donepezil, galantamine and rivastagimine. However, the tolerated doses of cholinesterase inhibitors, such as donepezil, produce only partial improvement in memory and can have severe gastrointestinal side-effects which can be problematic for patients that may have to take these drugs for prolonged durations. In the Merck study, investigators assessed the ability of PQCA (a positive allosteric modulator of M1 muscarinic receptors) to ameliorate cognitive deficits observed in aged APPSWE mice. They found that high doses of PQCA improved recognition memory just as well as donepezil in aged APPSWE mice. Interestingly, they found that administering sub-efficacious doses of PQCA and donepezil together significantly improved recognition memory, similar to the level produced by efficacious doses of donepezil by itself. Overall, the results in APPSWE mice concord with a long line of studies in rats and nonhuman primates highlighting the utility of M1 PAMs as another potential therapy for AD23.

These results not only highlight the therapeutic potential of M1 positive allosteric modulators, but also demonstrate the potential of using M1 PAMs, either in mono-therapy or in combination with the current standard of care, to achieve needed efficacy and escape the side-effects of acetylcholinesterase inhibitors.

A􀀀 accumulation is a key phenotype observed in APPSWE mice and is a target of many drug development efforts. Thus, the APPSWE mouse model maybe a suitable platform to address M1 muscarinic receptor based mechanistic questions in AD, and help accelerate the development of novel therapies that target aging associated degeneration and dementia in AD.

Tissue-engrafted models and their utility
The generation of human tissue-engrafted models has allowed scientists to not only recapitulate human organ systems in a physiological context, but they have also enabled recapitulation of clinical conditions. As a result, these models are widely used as preclinical platforms to evaluate the efficacy and safety of a variety of drugs.

The key to generating models with tissues or cells of human origin was the generation of mice with defective immune systems, which are reviewed in detail elsewhere24-27. One of the most immunodeficient mice is the NOG mouse as it lacks mature T, B, and NK cells, displays reduced complement activity and has dysfunctional macrophages and dendritic cells. This mouse has essentially become the ‘base’ for engraftment of a variety of human cells, tissue, and patient-derived tumours.

Immune system engrafted models and utility
Immune system engrafted mice are immunodeficient mice (eg, NOG or NSG mice) that have been implanted with various human immune system cell types. The human cells establish in the mouse and recapitulate various aspects of the human immune system. Given the vast complexity of the human immune system, it is important to note that while immune cells engrafted mice may mimic many functional aspects of the human immune system, no single model can capture the full complexity of human immunity. The NOG mouse can be engrafted with a variety of immune cell types such as hematopoietic stem cells (HSCs) for use in infectious disease research and oncology, or they can be engrafted with peripheral blood mononuclear cells (PBMC) to model disease involving dysfunctional T cells (eg, graft versus host disease).

In an example of the utility of such a mouse, NOG mice were engrafted with a human immune system by injection of CD34+ hematopoietic stem cells (HSCs). These engrafted mice were subsequently co-transplanted with a non-small cell lung cancer patient derived xenograft (PDX). This combined model was then used to test the efficacy of the immune checkpoint blocker ipilimumab, which is an anti-CTLA-4 mAb. Overall, the study showed that the PDX tumours grew in the immune engrafted mice, and that administration of ipilimumab caused T-cell proliferation and enhanced tumour infiltration by human CD3+ cells – the expected human immune system anti-tumour response elicited by ipilimumab.

This study highlights the utility of combining human immune system engrafted mice with PDX tumours to generate a preclinical platform that can be used to evaluate the efficacy of immunotherapies in oncology.

Liver engrafted models
Similar to immune system engrafted mice, immunodeficient mice can also be engrafted with functional human liver cells, and these models are increasingly being used in drug development studies28.

Areas such as DMPK, drug safety and infectious diseases, where the human liver plays a major role, stand to benefit immensely from these liver models. An exemplary challenge in the area of DMPK is where human hepatocytes are used in vitro to study metabolism and PK properties of drug candidates. However, as scientists have learned over the years, cells in a dish are not a great model for human liver and key aspects of drug disposition seen in an in vivo system, eg albumin binding of drugs or metabolite clearance systems, are often missed. By recapitulating the human liver in a physiological environment, liver engrafted models can provide a much needed in vivo alternative to in vitro human hepatocyte-based studies.

Several models are available for engraftment of human hepatocytes. The models essentially have an immunodeficient mouse as the base with an additional molecular mechanism to ablate the mouse hepatocytes so that they can be replaced with human hepatocytes28,29.

Briefly, these models include the scid/AlbuminuPA mouse30,31, the FRG mouse32, and the TKNOG mouse33. Each of the models differ primarily in how the mouse hepatocytes are ablated. For example, in the case of the TK-NOG mouse the Herpes Simplex Virus-1 Thymidine Kinsase (HSVtk) transgene expression is driven by the mouse albumin promoter to express in mouse liver cells. Treatment with the nucleoside analogue ganciclovir catalyses the conversion of the substrate into a toxic product that leads to the injury of mouse hepatocytes. Following ablation of the mouse hepatocytes, intrasplenic injection of human liver cells generates a mature and functional organ. The humanised liver in TK-NOG mice can be stably maintained for at least eight months and is capable of human-specific drug metabolism33.

A key advantage of using human liver-engrafted mice is that one can potentially engraft mice with different types of human hepatocytes representing different segments of the human population. These types of studies are helpful, for example, when addressing population differences in drug metabolism. Another major utility is in modelling infectious diseases such as Hepatitis B, Hepatitis C and malaria. Furthermore, human hepatocytes that have been used in vitro, can subsequently be engrafted into the mouse to create an in vivo system, which can help bridge the in vitro-in vivo gap which remains a major challenge in DMPK. Thus, the models may provide great value in predicting drug metabolism and disposition by combining in vitro and in vivo data.

Given the intricate and critical interplay between the liver and the human immune system, future iterations of these models are likely ones that can be engrafted with both human immune system and human liver cells34.

Precision-engineered mouse models have advanced greatly in the past few years and are filling the important knowledge gap that exists between mouse and human physiologies. These models are proving to be powerful tools to better help predict human outcomes before drugs are tested in humans. ‘Traditional’ preclinical models are largely inadequate when predicting human drug outcomes, with regard to efficacy and safety. This inadequacy undoubtedly contributes to late stage attrition of drug candidates thereby resulting in an increased cost of drug development. As more resources are designated on the development of more complex models, the future is likely to bring models that will continue to reduce late stage attrition of drug candidates. Given the intrinsic complexity of these models, selecting the right preclinical model will require careful consideration.

Dr Amar Thyagarajan is currently Senior Product Manager at Taconic. He manages several portfolios of Taconic’s genetically engineered models (GEMs) used in application areas such as neuroscience, ADME/Tox, cardiovascular disease and carcinogenicity testing. His technical background is in cell biology, neuroscience and chemical biology and he earned his PhD at the State University of New York, working on genetic regulatory networks involved in brain development and in neurodegenerative disorders.

1 Research/SpecialTopics/CriticalP athInitiative/CriticalPathOpport unitiesReports/ucm077262.htm.

2 Scheer, N, Snaith, M, Wolf, CR and Seibler, J 2013a. Generation and utility of genetically humanized mouse models. Drug Discov.Today. 18:1200-1211.

3 Shultz, LD, Ishikawa, F and Greiner, DL 2007. Humanized mice in translational biomedical research. Nat.Rev.Immunol. 7:118-130.

4 Shultz, LD, Brehm, MA, Garcia-Martinez, JV and Greiner, DL 2012. Humanized mice for immune system investigation: progress, promise and challenges. Nat.Rev.Immunol. 12:786-798.

5 Holzapfel, BM, Wagner, F, Thibaudeau, L, Levesque, JP and Hutmacher, DW 2015. Concise review: humanized models of tumor immunology in the 21st century: convergence of cancer research and tissue engineering. Stem Cells. 33:1696-1704.

6 Palmiter, RD, Norstedt, G, Gelinas, RE, Hammer, RE and Brinster, RL 1983. Metallothionein-human GH fusion genes stimulate growth of mice. Science. 222:809-814.

7 Scheer, N, Kapelyukh, Y, Rode, A, Oswald, S, Busch, D, McLaughlin, LA, Lin, Henderson, CJ, and Wolf, CR 2015. Defining Human Pathways of Drug Metabolism In Vivo through the Development of a Multiple Humanized Mouse Model. Drug Metab.Dispos.

8 Jaiswal, S, Sharma, A, Shukla, M, Vaghasiya, K, Rangaraj, N and Lal, J 2014. Novel preclinical methodologies for pharmacokinetic drug-drug interaction studies: spotlight on “humanized” animal models. Drug Metab.Rev. 46:475-493.

9 Hasegawa, M, Kapelyukh, Y, Tahara, H, Seibler, J, Rode, A, Krueger, S, Lee, DN, Wolf, CR and Scheer, N 2011a. Quantitative prediction of human pregnane X receptor and cytochrome P450 3A4 mediated drug-drug interaction in a novel multiple humanized mouse line. Mol.Pharmacol. 80:518-528.

10 International Transporter Consortium, Giacomini, KM, Huang, SM, Tweedie, DJ, Benet, LZ, Brouwer, KL, Chu, X, Dahlin, A, Evers, R, Fischer, V, Hillgren, KM, Hoffmaster, KA, Ishikawa ,T, Keppler, D, Kim, RB, Lee, CA, Niemi, M, Polli, JW, Sugiyama, Y, Swaan, PW, Ware, JA, Wright, SH, Yee, SW, Zamek-Gliszczynski, MJ and Zhang, L. Membrane transporters in drug development. Nat Rev Drug Discov. 2010 Mar;9(3):215-36.

11 Salphati, L, Chu, X, Chen, Prasad, LB, Dallas, S, Evers, R, Mamaril-Fishman, D, Geier, EG, Kehler, J, Kunta, J, Mezler, M, Laplanche, L, Pang, J, Rode, A, Soars, MG, Unadkat, JD, van Waterschoot, RA, Yabut, J, Schinkel, AH and N. Scheer, N 2014. Evaluation of organic anion transporting polypeptide 1B1 and 1B3 humanized mice as a translational model to study the pharmacokinetics of statins. Drug Metab.Dispos. 42:1301-1313.

12 Marone, PA, Hall, WC and Hayes, AW 2014. Reassessing the two-year rodent carcinogenicity bioassay: a review of the applicability to human risk and current perspectives. Regul.Toxicol.Pharmacol. 68:108-118.

13 Storer, RD, Sistare, FD, Reddy, MV and DeGeorge, JJ 2010. An industry perspective on the utility of short-term carcinogenicity testing in transgenic mice in pharmaceutical development. Toxicol.Pathol. 38:51-61.

14 Spalding, JW et al. Responses of transgenic mouse lines p53(+/-) and Tg.AC to agents tested in conventional carcinogenicity bioassays. Toxicol. Sci. 53, 213-23 (2000).

15 Jacobs, AC and Brown, PC. Regulatory Forum Opinion Piece*: Transgenic/Alternative Carcinogenicity Assays: A Retrospective Review of Studies Submitted to CDER/FDA 1997-2014. Toxicol Pathol. 2015 Jul;43(5):605-10.

16 Tamaoki, N. The rasH2 transgenic mouse: nature of the model and mechanistic studies on tumorigenesis. Toxicol. Pathol. 29 Suppl, 81-89 (2001).

17 Morton, D, Alden, C.L, Roth, AJ and Usui, T. The Tg rasH2 mouse in cancer hazard identification. Toxicol. Pathol. 30, 139-46 (2002).

18 Nambiar, PR and Morton, D 2013. The rasH2 mouse model for assessing carcinogenic potential of pharmaceuticals. Toxicol.Pathol. 41:1058-1067.

19 ICH. Concept Paper S1: Rodent Carcinogenicity Studies for Human Pharmaceuticals. (2012).

20 Federal Register. Food and Drug Administration, HHS. 2013. International Conference on Harmonisation; proposed change to rodent carcinogenicity testing of pharmaceuticals; request for comments. Notice; request for comments. Fed.Regist. 78:16681-16684.

21 Takanashi, S, Hara, K, Aoki, K, Usui, Y, Shimizu, M, Haniu, H, Ogihara, N, Ishigaki, N, Nakamura, K, Okamoto, M, Kobayashi, S, Kato, H, Sano, K, Nishimura, N, Tsutsumi, H, Machida, K and Saito, N 2012. Carcinogenicity evaluation for the application of carbon nanotubes as biomaterials in rasH2 mice. Sci.Rep. 2:498.

22 Urano, K, Suzuki, S, Machida, K, Sawa, N, Eguchi, N, Kikuchi, K, Fukasawa, K, Taguchi, F and Usui, T 2006. Use of IC tags in short-term carcinogenicity study on CB6F1 TGrasH2 mice. J.Toxicol.Sci. 31:407-418.

23 Puri, V, Wang, X, Vardigan, JD, Kuduk, SD and Uslaner, JM 2015. The selective positive allosteric M1 muscarinic receptor modulator PQCA attenuates learning and memory deficits in the Tg2576 Alzheimer’s disease mouse model. Behav.Brain Res. 287:96-99.

24 Flanagan, SP. 1966. ‘Nude’, a new hairless gene with pleiotropic effects in the mouse. Genet.Res. 8:295-309.

25 Ito, M, Hiramatsu, H, Kobayashi, K, Suzue, K, Kawahata, M, Hioki, K, Ueyama, Y, Koyanagi, Y, Sugamura, K, Tsuji, K, Heike, T and Nakahata, T 2002. NOD/SCID/gamma(c)(null) mouse: an excellent recipient mouse model for engraftment of human cells. Blood. 100:3175-3182.

26 Ishikawa, F, Yasukawa, M, Lyons, B, Yoshida, S, Miyamoto, T, Yoshimoto, G, Watanabe, T, Akashi, K, Shultz, LD and Harada, M 2005. Development of functional human blood and immune systems in NOD/SCID/IL2 receptor {gamma} chain(null) mice. Blood. 106:1565-1573.

27 Shultz, LD, Lyons, BL, Burzenski, LM, Gott, B, Chen, X, Chaleff, S, Kotb, M, Gillies, SD, King, M, Mangada, J, Greiner, DL and Handgretinger, R. 2005. Human lymphoid and myeloid cell development in NOD/LtSzscid IL2R gamma null mice engrafted with mobilized human hemopoietic stem cells. J.Immunol. 174:6477-6489.

28 Grompe, M and Strom, S 2013. Mice with human livers. Gastroenterology. 145:1209- 1214.

29 Scheer, N and Wilson, ID 2015. A comparison between genetically humanized and chimeric liver humanized mouse models for studies in drug metabolism and toxicity. Drug Discov Today. Sep 7. pii: S1359-6446(15)00342-6.

30 Rhim, JA, Sandgren, EP, Degen, JL, Palmiter, RD and Brinster, RL. Replacement of diseased mouse liver by hepatic cell transplantation. Science. 1994 Feb 25;263(5150):1149-52.

31 Tateno, C, Yoshizane, Y, Saito, N, Kataoka, M, Utoh, R, Yamasaki, C, Tachibana, A, Soeno, Y, Asahina, K, Hino, H, Asahara, T, Yokoi, T, Furukawa, T and Yoshizato, K. Near completely humanized liver in mice shows human-type metabolic responses to drugs. Am J Pathol. 2004 Sep;165(3):901-12.

32 Azuma, H, Paulk, N, Ranade, A, Dorrell, C, Al-Dhalimy, M, Ellis, E, Strom, S, Kay, MA, Finegold, M and Grompe, M 2007. Robust expansion of human hepatocytes in Fah-/- /Rag2-/-/Il2rg-/- mice. Nat.Biotechnol. 25:903-910.

33 Hasegawa, M, Kawai, K, Mitsui, T, Taniguchi, K, Monnai, M, Wakui, M, Ito, M, Suematsu, M, Peltz, G, Nakamura, M and Suemizu, H 2011b. The reconstituted ‘humanized liver’ in TK-NOG mice is mature and functional. Biochem.Biophys.Res. Commun.405:405-410.

34 Strick-Marchand, H, Dusséaux, M, Darche, S, Huntington, ND, Legrand, N, Masse-Ranson, G, Corcuff, E, Ahodantin, J, Weijer, K, Spits, H, Kremsdorf, D and Di Santo, JP. A novel mouse model for stable engraftment of a human immune system and human hepatocytes. PLoS One. 2015 Mar 17;10(3):e0119820.